Artículo Académico / Academic Paper
Recibido: 27-03-2023, Aprobado tras revisión: 14-06-2023
Forma sugerida de citación: Guañuna, G.; Chamba, M.; Granda, N.; Cepeda, J.; Echeverría, D.; Vargas, W. (2023). “Voltage
Stability Margin Estimation Using Machine Learning Tools”. Revista Técnica “energía”. No. 20, Issue I, Pp. 1-8
ISSN On-line: 2602-8492 - ISSN Impreso: 1390-5074
Doi: https://doi.org/10.37116/revistaenergia.v20.n1.2023.570
© 2023 Operador Nacional de Electricidad, CENACE
Voltage Stability Margin Estimation Using Machine Learning Tools
Estimación del Margen de Estabilidad de Voltaje Utilizando Herramientas de
Aprendizaje Automático
G.F. Guañuna1 M.S. Chamba3 N.V. Granda1
J.C. Cepeda1 D.E. Echeverría2 W. Vargas1
1Escuela Politécnica Nacional, Ecuador
E-mail: gabriel.guanuna@epn.edu.ec; nelson.granda@epn.edu.ec; jaime.cepeda@epn.edu.ec;
walter.vargas@epn.edu.ec
2Operador Nacional de Electricidad, CENACE
E-mail: decheverria@cenace.gob.ec
3CELEC EP Unidad de negocio Coca Codo Sinclair
E-mail: marlon.chamba@celec.gob.ec
Abstract
Real-time voltage stability assessment, via
conventional methods, is a difficult task due to the
required large volume of information, high
execution times and computational cost. Based on
this background, this technical work proposes an
alternative method for voltage stability margin
estimation through the application of artificial
intelligence and data mining algorithms. For this
purpose, 10 000 operate scenarios were generated
through Monte Carlo simulations, considering the
load variability and the n-1 security criterion.
Afterwards, the voltage stability margin of all
scenarios were determined using power-voltage (PV)
curves in order to obtain a database. This
information allowed structuring a data matrix for
training and evaluating an artificial neural network
and a support vector machine, capable of predicting
the voltage stability margin, even in real time. The
performance of the prediction tools was evaluated
through the mean square error and the coefficient of
determination. The proposed methodology was
applied to the IEEE 14 bus test system, showing so
promising results for both the neural network and
the vector machine, where the coefficients of
determination were 0.9153 and 0.8317, respectively.
Resumen
La evaluación de la estabilidad de voltaje en tiempo
real, mediante métodos convencionales, resulta en
una tarea difícil debido al gran volumen de
información, los elevados tiempos de ejecución y el
esfuerzo computacional requerido. Con estos
antecedentes, el presente trabajo técnico propone un
método alternativo que permite la estimación del
margen de estabilidad de voltaje a través de la
aplicación de algoritmos de inteligencia artificial y
minería de datos. Para ello, se generaron 10 000
escenarios operativos mediante simulaciones de
Monte Carlo, considerando la variabilidad de la
carga y el criterio de seguridad n-1. Posteriormente,
se determinaron los márgenes de estabilidad de
voltaje de todos los escenarios mediante el uso de las
curvas voltaje-potencia (PV, por sus siglas en inglés),
con la finalidad de obtener una base de datos. Esta
información permitió estructurar una matriz de
datos para entrenar y evaluar la red neuronal
artificial y la máquina vectorial de soporte, capaz de
predecir el margen de estabilidad de voltaje, incluso
en tiempo real. El desempeño de las herramientas de
predicción se evaluó a través del error cuadrático
medio y del coeficiente de determinación. La
metodología propuesta se aplicó al sistema de prueba
IEEE 14 bus, mostrando resultados prometedores
tanto para la red neuronal como para la máquina
vectorial, donde los coeficientes de determinación
fueron 0.9153 y 0.8317, respectivamente.
Index terms−− Voltage stability assessment, Monte
Carlo method, voltage stability margin estimation,
artificial intelligence algorithms.
Palabras clave−− Evaluación de la estabilidad de
voltaje, método de Monte Carlo, estimación del
margen de estabilidad de voltaje, algoritmos de
inteligencia artificial.
1
Edición No. 20, Issue I, Julio 2023
1. INTRODUCTION
The increase of electricity load as well as the
existence of new economic and environmental
constraints on generation dispatch and expansion of
transmission systems have caused the Electrical Power
System (EPS) to operate closer to its operating limits. In
these new operating conditions, the static and dynamic
security can be affected by voltage stability problems [1].
According to [2], voltage instability causes a progressive
and uncontrolled decrease of bus voltages after a
disturbance, a sudden increase in electrical load, a change
in system operating conditions or a combination of all of
them. Cepeda et al. established most of these events
occur in stressed EPS, where generators fail to maintain
normal voltage profiles at the busbars and transmission
lines are congested [3].
Currently, in the technical literature, several
approaches for voltage stability assessment have been
proposed: active power-voltage (PV) and reactive power-
voltage (Q-V) curves, modal analysis, sensitivity studies,
application of voltage stability indices and continuous
power flow (CPF), all of them for static analysis, as
shown by Patidar and Sharma [4]. However, from a real-
time perspective, such approaches require a large amount
of time and computational effort for the execution of
these methodologies. On the other hand, artificial
intelligence-based algorithms are the most important
tools to perform real-time static or dynamic security
monitoring and assessment of EPS by predicting voltage,
frequency and angle stability margins.
As demonstrated in [5], today, machine learning
(ML) based techniques such as artificial neural networks
(ANNs), decision trees (DTs), fuzzy logic (FL), adaptive
neuro-fuzzy inference system (ANFIS) and support
vector machines (SVMs) have become attractive tools for
solving nonlinear problems with desired speed and
accuracy. In particular, deep learning is used in [6], for
short-term voltage stability assessment of power systems
to learn the dependencies from post-disturbance system
dynamic trajectories. In this connection, it is important to
highlight that most of the current proposed
methodologies, oriented to use artificial intelligence-
based algorithms, have been applied to test power
systems, but their implementation to actual power
systems, together with proper contingencies
consideration continues to be scarce.
Based on these facts, this paper presents a novel
methodology based on artificial neural networks,
specifically multi-layer perceptron (MLP), and support
vector regression (SVR), to estimate the voltage stability
margin (VSM) using a validated database generated by
Monte Carlo simulations. The proposal is applied to the
IEEE 14 bus test power system.
The rest of the paper is organized as follows. A
theoretical review of voltage stability assessment
methodologies is presented in Section 2. Section 3
describes the proposed methodology that considers the
database generation, data processing and considerations
for machine learning training and testing. Moreover,
Section 4 shows the application example and obtained
results. Finally, the main conclusions are stated in
Section 5.
2. THEORETICAL REVIEW
2.1. Voltage Stability definition
According to IEEE (Institute of Electrical and
Electronics Engineers) / CIGRE (International Council
on Large Electric Systems), voltage stability refers to the
ability of a power system to maintain steady voltages at
all buses in the system after being subject to a disturbance
[7]. The phenomenon that occurs when the electric
system is unable to meet demand with steady voltages
under stress conditions is known as voltage instability.
According to [8], the factors contributing to voltage
stability are the generators’ reactive power limits, outage
of any equipment (transmission lines, generators or
transformers), load characteristics, characteristics of
reactive compensation devices and the action of voltage
control devices.
2.2. PV curves
PV curves are essential to analyze the voltage
stability of an EPS. They allow finding the critical
voltage instability point by increasing the power load
until the power flow does not converge (stability limit),
as shown by Amroune [9]. As demonstrated in [10], a
methodology is proposed to determine the voltage profile
power transfer limits of the monitored transmission
corridors using the Thevenin Equivalent method and the
determination of the PV curve in real-time. This allows
voltage stability assessment in real time and constitutes
an important basis for early-warning indicators.
However, this methodology assumes the availability of
phasor measurement units (PMU) at both sending and
receiving ends of the transmission corridor, which is not
always possible, Reddy et al. [11] and Lee and Han [12].
One of the most frequent terms related to voltage
instability is the voltage stability margin (VSM), which
corresponds to a measure of the distance from the initial
operating point to the critical point, as illustrated in Fig.1.
In the figure, voltages decay when there is an increase in
the transmitted active power. The voltage stability limit
is at the critical point (B), while the initial operating point
(A) corresponds to a less loaded state. The curve above
the critical point is known as the stable part, whereas the
rest of the curve is known as the unstable part. In
addition, if there is a change in the power factors of the
loads, the curves also change because a new operating
point of the system appears and thus a new voltage
stability limit, as shown by Patiño and Limas [13].
2
Guañuna et al. / Voltage Stability Margin Estimation Using Machine Learning Tools
According to Fig. 1, the VSM can be calculated using
the initial operating point and critical point as:
Where:
• 𝑃
𝑚𝑎𝑥: Final power of the loads in [MW].
• 𝑃
0: Base power of loads in [MW].
𝑴𝑬 =𝑷𝒎𝒂𝒙 − 𝑷𝟎
𝑷𝟎
(1)
Therefore, a high VSM value denotes a more stable
EPS, since it can transfer more power until the stability
limit is reached. On the other hand, according to [14], a
low index value indicates the power transfer is limited
because the system is more stressed, and therefore, the
system is more prompt to voltage instability. A secure
operation region definition, through the application of PV
curves, is useful to operators when taking preventive or
corrective measures in real time operation. In this sense,
this range of limits is subjective because it is related to
operating regulations, technical reports, operating
experience, among others.
Figure 1: PV curve example [13]
To give an example, the alert and alarm limits
associated with voltage stability are determined by the
approach adopted in [15], which shows how
measurements from distributed PMUs can be combined
with relevant transmission lines’ parameters, and be
handled to detect forthcoming voltage stability problems
in power systems at early stage.
In this paper, iterative power flow computations,
based on the PowerFactory PV curve analysis module, is
performed to create a VSM database [16].
2.3. Monte Carlo method as scenario generation tool
The Monte Carlo method allows, through successive
deterministic power flows, to solve probabilistic power
flows. The application of the Monte Carlo method for the
analysis of probabilistic power flows allows considering
the stochasticity of the system behavior with the purpose
of performing a study closer to reality, as demonstrated
in [17].
In stability studies, the Monte Carlo method has
allowed the generation of multiple probable scenarios to
determine system security indexes. In [18], a
methodology is proposed to assess the load uncertainty
impact on the transient stability of EPS based on
probabilistic analysis of the Critical Clearing Time.
Monte Carlo method has also been used to determine
different operating conditions to establish voltage
stability indexes of transmission lines.
In this paper, Monte Carlo simulation is applied to
perform iterative simulations oriented to determine the
VSM of several operating scenarios, including the n-1
security criterion. For this aim, the scripting capability of
PowerFactory is used to iteratively control the PV curve
analysis tool from Python.
2.4. Voltage stability assessment using machine
learning techniques
There are different approaches for voltage stability
analysis such as PV and QV curves, modal analysis,
voltage stability indices (VSI) and continuation power
flows (CPF). However, the application of these tools in
real time to large SEPs is inconvenient due to the
computational effort required by the high number of
iterations related to the methods. That said, alternative
approaches related to machine learning models (MLM)
need to be explored to overcome this computational
problem by interacting with technological tools, high-
level programming languages and data mining. The ML
includes many techniques such as artificial neural
networks (ANNs), decision trees (DTs), fuzzy logic (FL),
adaptive neuro-fuzzy inference system (ANFIS), support
vector machines (SVMs), among others.
Machine learning is a branch of artificial intelligence
that groups a set of methods for the creation of models
that learn from data with the purpose of making a
prediction or inference, as shown by Flach [19]. In this
regard, an approach to estimate the VSM using artificial
intelligence tools is presented in [20]. This methodology
applies voltage stability indexes (VSI) calculated from
synchrophasor measurements.
On the other hand, a new approach to estimate the
voltage stability margin through the combination
between a kernel extreme learning machine (KELM) and
a mean-variance mapping optimization (MVMO)
algorithm is presented in [21], where the Monte Carlo
method is employed to build the database for model
training and validation. A comprehensive review of the
application of machine learning tools such as artificial
neural networks (ANN), decision tree (DT), support
vector machines (SVM), for power system studies,
especially in cyber-attack detection, PQ perturbation
studies and dynamic security assessment studies is
presented in [22].
Machine learning algorithms are mainly split into two
groups:
3
Edición No. 20, Issue I, Julio 2023
2.4.1 Supervised algorithms
These algorithms use labeled data sets to create a
model that, using a vector of input and output features,
predicts the label of the feature vector. Regression and
classification are the two sub-groups associated with
supervised learning. In this regard, artificial neural
networks and support vector machines are used in this
paper in their regression versions for predicting the VSM
of the power system.
2.4.2 Unsupervised algorithms
These algorithms use an unlabeled dataset to find a
final structure in the data, using only one set of inputs, as
shown Echeverría [23]. The main purposes odd these
algorithms are the data dimensionality reduction and
clustering. In this connection, the principal component
analysis (PCA) is used in this paper.
3. METODOLOGY
This technical work estimates the voltage stability
margin of an EPS using machine learning tools and a
database generated by Monte Carlo simulations. For this
purpose, Python programming language and
PowerFactory software are used. In the first stage, a
validation of PowerFactory PV curves module using
Matpower is performed. The stage two consists of
database generation, data processing and machine
learning algorithms application. Finally, VSM estimation
and results analysis are performed in stage three. Fig. 2
schematizes the proposed methodology.
Figure 2: Methodology stages for estimating the voltage stability
margin
3.1. Stage 1: Matpower and PowerFactory
The validation of PowerFactory PV curves module
using Matpower allows to establish the theoretical-
technical support to implement the methodology of this
study. This comparison is performed to verify how close
to the nose of PV curves can be reached by the algorithm
implemented in PowerFactory since it does not exactly
accomplish CPF formal theory, whereas Matpower does.
The standard IEEE 14 bus test system is used as a case
study for this purpose.
3.2. Stage 2: Software development
The technical study aims to estimate the voltage
stability margin from a validated database, considering
machine learning algorithms. The following sections
present the general procedure developed for database
generation, data processing and considerations for the
implemented algorithms.
3.2.1 Database generation
During the implementation of the simulation
proposal, two processes are taken into account. The first
process consists of generating the operational scenarios
through communication between Python and
PowerFactory. This process uses Monte Carlo method to
generate multiple operating scenarios considering the
variability of the load. For this purpose, optimal power
flows (OPF) are first executed by PYPOWER. According
to [24], PYPOWER is a power flow and OPF solver,
which is a port of MATPOWER to the Python
programming language.
On the one hand, the Monte Carlo method allows
considering the uncertainty of the demand, while the
optimal power flows are used to obtain a proper dispatch
of the generation units. It should be noted that the use of
OPF in the face of load variations allows solving the
problem associated with congestion of transmission lines
near the slack generator when only power flows are used.
Figure 3: Flowchart for the generation of operational states
Fig. 3 shows the procedure adopted for the generation
of operating states, where the input data are: the operating
states in DIgSILENT PowerFactory, the generating costs
of each unit and the stochasticity of each system load
represented by probability density functions (PDF).
4
Guañuna et al. / Voltage Stability Margin Estimation Using Machine Learning Tools
On the other hand, the second process allows the
calculation of the VSM from the PV curves obtained
from each operating scenario, generated by Monte Carlo
simulations. These stability margins are arranged in a
large matrix relating the main system variables. In
addition, a set of transmission lines is chosen for each
system under study, and in this way, through
programming, the "n-1" criterion is assessed at the time
of executing a power flow or the PV curves module. In
this connection, all the information, including the
stability margin, is stored in pre-contingency and post-
contingency data sets. The execution of this procedure
developed in Python can be done directly in
PowerFactory or through an external connection known
as "engine mode", which allows the program to be
controlled without the need for it to be open.
3.2.2 Data proccesing
The data processing is performed in the Python
programming environment, due to its versatility when
handling variables and the extensive documentation with
respect to machine learning and data mining models. In
this sense, the whole set of data obtained is structured and
debugged to obtain the pre-contingency and post-
contingency matrices, considering the individuals or
samples by the number of rows, while, the features by the
number of columns. The number of rows is set to 10,000
samples and the number of columns depends on the
system or region of analysis (the features are the set of
electrical variables that reflect the system steady-state of
each operating scenario). After this, a reduction of the
matrix dimensions is performed by means of PCA,
considering that, in order to reduce the number of
features.
Figure 4: Flow chart for the implementation of the established
models
Once the files containing all the information from the
base case, pre-contingency case and post-contingency
case have been created, the data are processed so that they
can be used by the machine learning algorithms and the
models performance is evaluated, consequently. Fig. 4
shows the program flow chart for data structuring and
processing.
3.2.3 Considerations
Artificial neural networks and support vector
machines available in the "Scikit-learn" library (machine
learning and data analysis library developed in Python
programming language) are trained and implemented.
The hyperparameters inside each model can be modified
according to the specific documentation. It should be
emphasized that, during the modeling of the regressors,
the input variables are found in the pre-contingency
matrix and the output variables are found in the post-
contingency matrices, this approach will allow to
properly perform the training and validation of the
regressors. In addition, hyperparameters optimization of
each model is performed using GridSearchCV, as shown
Predregosa et al. [25].
3.3. Stage 3: Results
Finally, the obtained results are analyzed by
calculating the mean square error (MSE) and the
coefficient of determination (R2) for the purpose of
assessing the performance of each regressor. According
to [26], MSE shows the average squared difference
between the obtained and predicted values. A value close
to zero of MSE indicates that the model fits the data set
properly. R2, on the other hand, quantifies the linear
relationship between the obtained value and the predicted
value. A value close to “one” indicates that the model
presents an appropriate fit. The entire process is
performed using cross-validation.
4. RESULTS – APPLICATION TO THE IEEE 14
BUS SYSTEM
The implemented algorithms are assessed through an
optimization of the hyperparameters of each model,
allowing to verify how well they fit the database. In
addition, eight transmission lines are selected based on
contingency analysis, to assess the “n-1” criterion. In this
case, eight transmission lines were considered to
calculate the VSM and to obtain the information of the
pre and post-contingency system variables. For this
reason, eight regressors were used. However,
performance index results are only presented when
transmission line “6-13” is out of service.
Fig. 5 shows the reduction of the voltage stability
margin of the critical bus of the system, before and after
the occurrence of the contingency. This demonstrates the
capability of the system to adjust to the active and
reactive power requirements.
5
Edición No. 20, Issue I, Julio 2023
Figure 5: Example of PV curves before and after the L/T 6-13
operation exit
Table 1 shows the results for both the artificial neural
network and the support vector machine. In this regard,
for SVM, MSE gets worse when the optimization is
applied since the error increases, while, R2 index, also
gets worse due to the fact that it moves away from one.
For these reasons it can be said that SVM does not
correctly fit to the data set. Whereas, for ANN, MSE
improves when the optimization is applied since the error
is reduced, while, R2 index, also improves because it is
closer to unity. As a conclusion, ANN approach fits
better to the data set.
Table 1: Resultados de los índices de rendimiento - L/T 6-13
Models \ Indexes
MSE
R2
ANN
0.07436
0.89741
SVM
0.07637
0.90199
ANN - GridSearchCV
0.06599
0.91531
SVM - GridSearchCV
0.13115
0.83171
The implemented models are adjusted depending on
the established data set, i.e., the results obtained when the
SVM is applied to the test system are not so efficient
since it worsens the performance indexes. This can be
justified by the outliers and the behavior of the system
when calculating the stability margin by means of the PV
curves. With this in mind, analyzing the case study, it is
concluded that the best machine learning algorithm is the
artificial neural network, since it presents a better
prediction of the stability margin VSM post-contingency
VSM pre-contingency before and after the optimizer. In
addition, the proposed methodology was applied to the
Ecuadorian National Interconnected System,
demonstrating the robustness of the application and the
improvement of performance indexes.
5. CONCLUSIONS
The machine learning model structuring requires to
define the percentage of data to be used for training,
validation and testing of the model. For the present case,
80% was for training and 20% for model evaluation.
However, the use of validation data set causes a
considerable loss of samples which affects machine
learning, in that sense, cross-validation method was
implemented to take advantage of the largest amount of
data for the final assessment of the model.
The neural network and the support vector machine
present adequate performance indexes to the voltage
stability margin prediction. In particular, when the
optimizer was used, R2 for ANN was 0.9153, and for
SVM was 0.8317. Similar R2 results were obtained when
the optimizer was not applied. However, when the
optimizer was used, MSE for ANN was 0.0659, and for
SVM was 0.1311. Therefore, the artificial neural network
has the best prediction and therefore it is the regressor
that best fits the data set of the tackled problem.
The load variability was performed through a
percentage change in the loads, however, it is
recommended to use daily load curves (industrial,
commercial and residential) to know the real load
behavior in a 24-hour time interval.
REFERENCIAS BIBLIOGRÁFICAS
[1] J. Rueda et al., “Dynamic Vulnerability Assessment
and Intelligent Control for Sustainable Power
Systems,” Wiley-IEEE Press, 2018.
[2] C. Andersson, J. E. Solem, and B. Eliasson,
“Classification of power system stability using
support vector machines,” 2005 IEEE Power Eng.
Soc. Gen. Meet., vol. 1, no. 2, pp. 650–655, 2005,
doi: 10.1109/pes.2005.1489266.
[3] N. P. Patidar and J. Sharma, “Loadability margin
estimation of power system using model trees,” 2006
IEEE Power India Conf., pp. 338–343, 2005, doi:
10.1109/POWERI.2006.1632534.
[4] M. Bilgen and S. Ozdemir, “Comparison of Real-
Time Voltage Stability Assessment Methods,” 13th
Int. Conf. Electr. Electron. Eng., pp. 73–77, 2021,
doi:10.23919/ELECO5474.2021.967773 4.
[5] A. Adhikari, S. Naetiladdanon, A. Sagswang, and S.
Gurung, “Comparison of voltage stability
assessment using different machine learning
algorithms,” 2020 IEEE 4th Conf. Energy Internet
Energy Syst. Integr. Connect. Grids Towar. a Low-
Carbon High-Efficiency Energy Syst. EI2 2020, pp.
2023–2026, 2020, doi:
10.1109/EI250167.2020.9346750.
[6] M. Zhang, J. Li, Y. Li, and R. Xu, “Deep Learning
for Short-Term Voltage Stability Assessment of
Power Systems,” IEEE Access, vol. 9, pp. 29711–
29718, 2021.
[7] N. Hatziargyriou et al., “Definition and
Classification of Power System Stability - Revisited
& Extended,” IEEE Trans. Power Syst., vol. 36, no.
4, pp. 3271–3281, 2021, doi:
10.1109/TPWRS.2020.3041774.
6
Guañuna et al. / Voltage Stability Margin Estimation Using Machine Learning Tools
[8] M. Amroune, “Machine Learning Techniques
Applied to On-Line Voltage Stability Assessment: A
Review,” Arch. Comput. Methods Eng., vol. 28, no.
2, pp. 273–287, 2021, doi: 10.1007/s11831-019-
09368-2.
[9] P. Kundur, Power System Stability And Control.
New York: McGraw-Hill, 1994.
[10] J. Cepeda, P. Verdugo, and G. Argüello, “Monitoreo
de la Estabilidad de Voltaje de Corredores de
Transmisión en Tiempo Real a partir de Mediciones
Sincrofasoriales,” Rev. EPN, vol. 33, no. 3, 2014.
[11] A. Reddy, K. Ekmen, V. Ajjarapu, and U. Vaidya,
“PMU based real-time short term voltage stability
monitoring - Analysis and implementation on a real-
time test bed,” North Am. Power Symp. NAPS,
2014, doi: 10.1109/NAPS.2014.6965485.
[12] Y. Lee and S. Han, “Real-time voltage stability
assessment method for the Korean power system
based on estimation of Thévenin equivalent
impedance,” Appl. Sci., vol. 9, no. 8, 2019, doi:
10.3390/app9081671.
[13] G. F. Patiño and G. A. Limas, “Metodologías para el
análisis de estabilidad de tensión en estado
estacionario,” B.S. thesis, Univ. Tecnológica de
Pereira, 2008.
[14] L. Chiza and J. Cepeda, “Predicción del Margen de
Estabilidad de Corredores de Transmisión
Aplicando Criterios de Minería de datos y
Algoritmos de Machine Learning,” Rev. Técnica
“energía,” vol. 18, no. 1, pp. 37–47, 2021, doi:
10.37116/revistaenergia.v18.n1.2021.466.
[15] Y. Narcisse and N. Tchokonte, “Real time
identification and monitoring of the voltage stability
margin in electric power transmission systems using
synchronized phasor measurements,” Ph.D.
dissertation, Univ. Kassel, Germany, 2009.
[16] DIgSILENT GmbH. PowerFactory 2021 - User
Manual. (2021). Accessed: Oct. 16, 2022.
[Online].Available: https://www.digsilent.de/en/
downloads.html
[17] B. Marah and A. O. Ekwue, “Probabilistic load
flows,” Proc. Univ. Power Eng. Conf., vol. 2015-
Novem, no. 1, 2015.
[18] M. S. Chamba, W. A. Vargas, and J. Cristobal
Cepeda, “Stochastic assessment and risk
management of transient stability based on
powerfactory and python interface,” IEEE PES
Transm. Distrib. Conf. Exhib. - Lat. Am. T D LA,
2020.
[19] P. Flach, Data, Machine Learning: The Art and
Science of Algorithms that Make Sense of.
Cambridge University Press, 2012.
[20] K. D. Dharmapala, A. Rajapakse, K. Narendra, and
Y. Zhang, “Machine Learning Based Real-Time
Monitoring of Long-Term Voltage Stability Using
Voltage Stability Indices,” IEEE Access, vol. 8, pp.
222544–222555, 2020.
[21] W. M. Villa-Acevedo, J. M. López-Lezama, and D.
G. Colomé, “Voltage stability margin index
estimation using a hybrid kernel extreme learning
machine approach,” Energies, vol. 13, no. 4, 2020,
doi: 10.3390/en13040857.
[22] O. A. Alimi, K. Ouahada, and A. M. Abu-Mahfouz,
“A Review of Machine Learning Approaches to
Power System Security and Stability,” IEEE Access,
vol. 8, pp. 113512–113531, 2020, doi:
10.1109/ACCESS.2020.3003568.
[23] D. E. Echeverría, “Evaluación y mejora de la
estabilidad transitoria de sistema eléctricos en
tiempo real utilizando PMUs,” Ph.D. dissertation,
Univ. Nacional de San Juan, 2021.
[24] R. Lincoln, Pypower Documentation (2017).
[Online]. Available: https://rwl.github.io/PYPO
WER/PYPOWER.pdf
[25] F. Predregosa et al., “Scikit-learn: Machine Learning
in Python,” J. Mach. Learn. Res., vol. 12, pp. 2825–
2830, 2011, [Online]. Available: https://scikit-
learn.org/stable/modules/cross_validation.html#cro
ss-validation.
[26] S. Kokoska and D. Zwillinger, Standard Probability
and Statistics Tables and Formulae. New York,
2000.
Gabriel F. Guañuna.- Nació en la
ciudad de Quito - Ecuador, en 1997.
Cursó sus estudios secundarios en
Quito, en el Colegio Técnico
Salesiano Don Bosco Kennedy,
dónde obtuvo el título de bachiller
técnico en Instalaciones, Equipos y
Máquinas Eléctricas. Obtuvo su
título de Ingeniero Eléctrico en 2022 en la Escuela
Politécnica Nacional. Es miembro voluntario en, IEEE
Student Branch Escuela Politécnica Nacional en el
capítulo Power & Energy Society. Áreas de Interés:
Estabilidad de voltaje, redes inteligentes y operación de
sistema eléctricos de potencia, machine learning.
Marlon S. Chamba.- Nació en
Loja, Ecuador, en 1982. Obtuvo su
título de Ingeniero Eléctrico en
2007 en la Escuela Politécnica
Nacional, Quito-Ecuador. Realizó
su doctorado en el Instituto de
Energía Eléctrica, Universidad
7
Edición No. 20, Issue I, Julio 2023
Nacional de San Juan, San Juan, Argentina, favorecido
con una beca del Servicio Alemán de Intercambio
Académico (DAAD). Finalmente obtuvo el título de
Doctor en Ingeniería Eléctrica en 2016. Sus campos de
interés especiales comprenden el control y la estabilidad
de los sistemas de energía en tiempo real, tecnología de
medición sincrofasorial, los sistemas de monitoreo de
área amplia, el desarrollo de redes inteligentes, la
confiabilidad de los sistemas de energía y el despacho
económico de energía.
Nelson V. Granda.- Obtuvo el
título de Ingeniero Eléctrico en la
Escuela Politécnica Nacional en el
año 2006 y de Doctor en Ciencias
de la Ingeniería Eléctrica en la
Universidad Nacional de San Juan
(Argentina), en el año 2015. Se ha
desempeñado como Ingeniero
Eléctrico en varias instituciones del sector eléctrico y
petrolero como son el Operador Nacional de Electricidad
(CENACE), Petroamazonas EP y CELEC-EP
TRANSELECTRIC. Actualmente, se desempeña como
parte del staff docente del Departamento de Energía
Eléctrica de la Escuela Politécnica Nacional. Sus áreas de
interés son análisis y control de sistemas eléctricos de
potencia en tiempo real y aplicaciones de Sistemas de
Medición de Área extendida (WAMS) basados en
unidades de medición sincrofasorial (PMU).
Jaime C. Cepeda.- Nació en
Latacunga, Ecuador en 1981.
Obtuvo el título de Ingeniero
Eléctrico en la Escuela Politécnica
Nacional EPN en 2005, y obtuvo su
doctorado en Ingeniería Eléctrica
en el Instituto de Energía Eléctrica
de la Universidad Nacional de San
Juan, San Juan, Argentina. Además, obtuvo el título de
Máster en Big Data por la Universidad Europea Miguel
de Cervantes, Valladolid, España en 2021. Actualmente
se desempeña como profesor universitario a tiempo
completo en programas de Maestrías y Doctorado en la
EPN. Sus campos de interés especial comprenden el
modelado de sistemas de potencia, la evaluación de la
seguridad, la tecnología de medición sincrofasorial, el
monitoreo de área amplia, los sistemas de protección y
control, y la aplicación de técnicas de inteligencia
computacional en el análisis de los sistemas de potencia.
Diego E. Echeverría.- Obtuvo su
título de Ingeniero Eléctrico en
2006 en la Escuela Politécnica
Nacional, Quito-Ecuador. Realizó
su doctorado en el Instituto de
Energía Eléctrica, Universidad
Nacional de San Juan, San Juan,
Argentina, favorecido con una beca
del Servicio Alemán de Intercambio Académico
(DAAD). Obtuvo el título de Doctor en Ingeniería
Eléctrica en diciembre de 2021. Actualmente trabaja en
Ecuador en el Operador Nacional de Electricidad
CENACE como Subgerente Nacional de Investigación y
Desarrollo. Sus campos de interés especiales
comprenden el control y la estabilidad de los sistemas de
energía en tiempo real, la tecnología de medición
sincrofasorial, los sistemas de monitoreo de área amplia
y el desarrollo de redes inteligentes.
Walter Vargas.- Nació en
Guayaquil, Ecuador en 1984-
Recibió sus títulos de Ingeniero en
Electricidad, especialización
Potencia (2007) en la Escuela
Superior Politécnica del Litoral y el
de Máster en Sistemas de Energía
Eléctrica (2013) en la Universidad
de Sevilla. Entre 2013 y el 2017 trabajó en la sección de
Estudios Eléctricos del Departamento de Centro de
Operaciones de CELEC EP – Transelectric. Actualmente
se desempeña como profesor universitario a tiempo
completo en la EPN. Sus áreas de interés incluyen la
optimización, confiabilidad, evaluación de
vulnerabilidad en tiempo real y el desarrollo de Smart
Grids.
8
